首页|Data on Computational Intelligence Detailed by Researchers at Guangzhou University (Effective Single-step Adversarial Training With Energy-based Models)
Data on Computational Intelligence Detailed by Researchers at Guangzhou University (Effective Single-step Adversarial Training With Energy-based Models)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning DailyNews Daily News - New research on Machine Learning - Com putational Intelligence is the subject of areport. According to news reporting originating in Guangzhou, People’s Republic of China, by NewsRxjournalists, res earch stated, “Adversarial training (AT) is one of the most effective ways again st adversarialattacks. However, multi-step AT is time-consuming while single-st ep AT is ineffective.”Financial support for this research came from National Natural Science Foundatio n of China (NSFC).The news reporters obtained a quote from the research from Guangzhou University, “In this paper, wepropose an Energy-AT framework to make single-step AT as eff ective as multi-step ones, by exploiting thetwo properties of energy-based mode ls (EBM). First, we utilize the Helmholtz free energy in EBM to pushgenerated e xamples to be outside of the distribution boundaries of their categories, such t hat they are moreadversarial. Second, we apply an adaptive temperature scheme i n EBM to amplify the training gradientsof weak adversarial examples targetedly, such that those originally hard-to-learn examples contribute to therobustifica tion of models also.”
GuangzhouPeople’s Republic of ChinaAsiaComputational IntelligenceMachine LearningGuangzhou University
NETL
NSTL
